Re: [Mldonkey-users] List of 'nice to have's and bugs

[Markus Hitter]

Am 22.11.2003 um 14:27 schrieb Pierre Etchemaite:

> You're fighting against the fact that you have less interesting 
> sources to get the few chunks that you're missing to complete a file 
> against all the sources available to resume a file that's still far 
> from being completed.

Of course. As incomplete files are useless and far from complete files 
are more easily to continue, this makes preferring almost complete 
files even more urgent.

> Check donkeyOneFile.find_client_block, it only switches to another 
> file when nothing interesting can be selected for current file.

... which is wrong, IMHO.

Well, I can't find a file with this name, but anyway. Common scenario 
is, as sources drop in, you first get a chance to download a chunk for 
a more incomplete file. With current behaviour, mldonkey whould 
continue with this file until it is completed or there are no more 
chunks for this file available. A file with a high priority or an 
almost complete file is left untouched. Even raising the priority of a 
file while mldonkey is running has no effect at all. This matches 
exactly my observations.

A better algorithm whould check priorities/completeness of all files at 
each chance, the decision which chunk to request next, is to be made. 
This algorithm whould continue with low priority / very incomplete 
files if there are no chunks for files with high priority available, 
only.


Enjoy,
Markus



P.S.: Sorry, I can't come up with a patch, I'd have to learn ocaml, 
first.

- - - - - - - - - - - - - - - - - - -
Dipl. Ing. Markus Hitter
http://www.jump-ing.de/